Medical image segmentation method based on adaptive anisotropic convolution

Through the methods of adaptive anisotropic convolution and cross-scale feature fusion, the problem of insufficient accuracy of deep learning in kidney tumor segmentation is solved, and efficient segmentation of small-scale targets is achieved, which is suitable for clinical applications.

CN120726076AActive Publication Date: 2025-09-30NANCHANG CAMPUS OF EAST CHINA UNIV OF TECH
View PDF 5 Cites 7 Cited by

Patent Information

Application Number
CN202511244241.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-02
Publication Date
2025-09-30
Estimated Expiration
2045-09-02

AI Technical Summary

Technical Problem

Existing deep learning methods have difficulty in effectively capturing global information in the task of kidney tumor segmentation, especially the segmentation accuracy of small-scale targets with complex shapes is insufficient, and there are deficiencies in multi-scale feature adaptation and fine-grained segmentation.

Method used

Adaptive anisotropic convolutional layers are used to extract feature information in different directions in parallel, and adaptive attention mechanisms are used to dynamically allocate weights. Combined with cross-scale feature fusion and multi-stage deep supervision, model parameters are optimized to improve segmentation accuracy.

Benefits of technology

It significantly improves the segmentation accuracy of small-scale targets, especially the segmentation effect of complex organs such as kidney tumors, simplifies the processing flow, and is suitable for clinical practice applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120726076A_ABST
    Figure CN120726076A_ABST
Patent Text Reader

Abstract

The invention provides a medical image segmentation method based on adaptive anisotropic convolution, and the method comprises the steps: obtaining a three-dimensional medical CT data set comprising images and labels of a plurality of abdominal organs and kidney tumors, and carrying out the preprocessing of the data set; dividing a data set into a training set and a test set for model training and evaluation; designing a three-dimensional medical image segmentation network model based on an adaptive anisotropic convolutional layer, and inputting the preprocessed training set into the three-dimensional medical image segmentation network model, the three-dimensional medical image segmentation network model is trained through parallel multi-modal convolution, adaptive attention weight generation, weighted feature dynamic fusion and multi-stage deep supervision, and model parameters are optimized; and applying the optimized three-dimensional medical image segmentation network model to a test set, generating a three-dimensional segmentation result with clear boundary and complete reserved details, and providing support for clinical diagnosis and treatment planning.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Image classification method based on convolution neural network

    CN107341518A

  • Medical image segmentation device based on global information perception

    CN116258933A

  • Medical image automatic diagnosis method and system based on deep learning

    CN120340784A

  • Medical image semantic segmentation method based on attention mechanism optimization

    CN120496757A

  • Attention-based joint image and feature adaptive semantic segmentation method

    US20230281828A1

Cited By

  • Medical image intelligent processing method and system

    CN121190496A

  • Muscle segmentation method based on state space model and boundary perception attention

    CN121213594A

  • A muscle segmentation method based on state space model and boundary-aware attention

    CN121213594B

  • Operation area lesion image segmentation method based on multi-scale dynamic segmentation kernel

    CN121437540A

  • A surgical area lesion image segmentation method based on a multi-scale dynamic segmentation kernel

    CN121437540B